# Appendix F: Case law and authority

The cases and rules this book is built on, briefly. Each summary is for orientation; consult primary sources for any operational use.

## *United States v. Kovel*, 296 F.2d 918 (2d Cir. 1961)

The foundational case. A tax attorney engaged an accountant to help interpret a client's books. The grand jury subpoenaed the accountant. The Second Circuit, in an opinion by Judge Henry Friendly, held that the privilege extends to communications with a non-lawyer agent retained by counsel to assist in providing legal advice. Friendly's translator analogy, the agent as someone who lets the lawyer understand what the client is saying, has been the operative metaphor for sixty-five years and is the basis for extending privilege to AI tools used at counsel's direction.

## *Upjohn Co. v. United States*, 449 U.S. 383 (1981)

The modern statement of attorney-client privilege in federal practice. The Supreme Court rejected the narrow control-group test and held that privilege extends to communications by employees within a corporate structure, when made for the purpose of obtaining legal advice. Together with *Hickman*, *Upjohn* establishes the strong protection of opinion work product. The case is frequently cited in privilege challenges and is the standard authority for the elements of the privilege.

## *Hickman v. Taylor*, 329 U.S. 495 (1947)

The origin of the work-product doctrine. The Supreme Court held that materials prepared in anticipation of litigation by counsel or counsel's representative are protected from discovery, with mental impressions, conclusions, and legal theories receiving the strongest protection. AI-assisted analyses prepared at counsel's direction map directly onto the work-product category.

## *United States v. Heppner*, S.D.N.Y. No. 25-cr-00503-JSR (Feb. 17, 2026)

The case that occasioned this book. On the Government's motion — not a defense motion to suppress — Judge Rakoff ruled that thirty-one documents Bradley Heppner generated with a consumer AI platform were protected by neither the attorney-client privilege nor the work product doctrine. The court found they lacked at least two, and possibly all three, elements of the privilege: the platform is not an attorney, which alone disposed of the claim; the consumer privacy policy permitted training on inputs and outputs and disclosure to third parties, so the communications were not confidential; and they were not made to obtain legal advice from the platform. In one hedged sentence of dictum the court observed that had counsel directed the use, the platform "might arguably" have functioned as a lawyer's agent within the privilege. That sentence is the conditions this book is written around. It is not a holding, and it addresses only one of the three defects.

## *In re Grand Jury Subpoena Dated March 9, 2001*, 179 F. Supp. 2d 270 (S.D.N.Y. 2001)

An extension of the Kovel doctrine to a public-relations consultant retained by counsel in a criminal matter. Useful authority for the proposition that the Kovel doctrine is not limited to technical experts and applies broadly to agents retained to facilitate legal advice. Strengthens the analogy to AI as a Kovel agent.

## *Cavallaro v. United States*, 284 F.3d 236 (1st Cir. 2002)

A clarifying limit on Kovel. The First Circuit held that the agent's role must be to facilitate communication between the attorney and client, not merely to provide independent expert advice. The case is operationally important for framing the engagement: AI is engaged *by the attorney* to assist in rendering advice, not as an independent advisor. The distinction between *agent* and *independent expert* is the doctrinal hinge.

## Authorities for the offense chapter

The material in chapters 11 and 12 rests on a different set of authorities than the privilege chapters. These are the ones to have at hand.

### *The New York Times Co. v. OpenAI, Inc.* (S.D.N.Y. 2025)

In May 2025 the magistrate judge entered a preservation order requiring OpenAI to retain output log data it would otherwise have deleted, overriding user deletion requests and the company's own retention policy. The order was narrowed in September 2025 so that it no longer reached all new logs going forward, and the plaintiffs' access demand was ultimately negotiated to a sample of roughly twenty million conversations. The case is a copyright dispute, but its operational lesson is general: user-side deletion is a client-side gesture, and a court that wants the underlying records can reach them.

### Stored Communications Act, 18 U.S.C. §§ 2701–2712

Section 2702(a) prohibits a provider of electronic communication service or remote computing service to the public from divulging the contents of stored communications. Courts have consistently held that a civil subpoena is not an exception, which is why a subpoena to an AI provider for conversation content will draw an objection rather than a production. Non-content records, subscriber information, and account existence remain reachable. Get content from the party; get proof of the account from the provider.

### Fed. R. Civ. P. 34(a)(1) — possession, custody, or control

The rule reaches material a responding party has the legal right or practical ability to obtain. Every major AI product offers the account holder a self-service export that delivers a complete archive on demand, which makes the control argument unusually clean. Cite the product's own help documentation.

### Fed. R. Civ. P. 37(e) — failure to preserve ESI

Where ESI that should have been preserved is lost because a party failed to take reasonable steps, the court may order curative measures on a finding of prejudice, and on a finding that the party acted with intent to deprive another party of the information may presume the information was unfavorable, instruct the jury accordingly, or dismiss. A preservation letter that names AI tools specifically is what converts later deletion from an oversight into an intent question.

### Fed. R. Civ. P. 26(b)(1) and 26(f)

Proportionality governs the scope of AI-conversation discovery and is the ground on which overbroad requests get cut back. Rule 26(f) is the conference at which AI data should be named in the ESI protocol; a generic reference to electronic communications will not resolve a later dispute about whether chats were covered.

### ABA Formal Opinion 512 (July 2024)

Addresses lawyers' use of generative AI under the Model Rules, working through competence (Rule 1.1 and comment 8), confidentiality (Rule 1.6, including the reasonable-efforts obligation in 1.6(c)), communication (Rule 1.4), fees, and supervision (Rules 5.1 and 5.3). Read together with Rule 1.4, it is the strongest available support for the proposition in chapter 12 that a lawyer who knows a client is using consumer AI on the matter has something to say about it. State bar opinions in California, Florida, New York, New Jersey, and elsewhere reach compatible conclusions; check your own jurisdiction.